International Journal of Drug Delivery Technology
Volume 16, Issue 3s, 2026

Predictive Analytics in Precision Drug Delivery: A Data-Driven Approach Using Machine Learning

Dr. Geeta Salunke 1, Dr. Shweta Sadanand Salunkhe 2, Prof. Kanchan Mahajan 2, Dr. Mamta Sanjay Koban 3, Dr. Manish Rana 4, Dr. Altaf Osman Mulani 5

1Assistant Professor, Department of Electronics & Telecommunication Engineering, AISSMS College of Engineering, Pune, Maharashtra, India.
Email: geetasalunke@gmail.com

2Assistant Professor, Department of Electronics & Telecommunication Engineering, Bharati Vidyapeeth's College of Engineering for Women, Pune, Maharashtra, India.
Email: shweta.salunkhe@bharatividyapeeth.edu; kanchan.mahajan@bhartividyapeeth.edu

3Assistant Professor, Department of Electronics & Telecommunication Engineering, AISSMS Institute of Information Technology, Pune, Maharashtra, India.
Email: mamta.wanjre@aissmsioit.org

4Associate Professor, Shah and Anchor Kutchhi College of Engineering, Mumbai, Maharashtra, India.
Email: manishrana23@gmail.com

5Professor, Department of Electronics & Telecommunication Engineering, SKN Sinhgad College of Engineering, Pandharpur, Maharashtra, India.
Email: draomulani.vlsi@gmail.com


ABSTRACT

Precision drug delivery represents the convergence of pharmacogenomics, physiologically-based pharmacokinetic (PBPK) modeling, and advanced machine learning to individualize drug therapy based on patient-specific biological, genetic, and clinical characteristics. Despite substantial advances in personalized medicine, therapeutic drug monitoring and dose optimization in complex patient populations—including the elderly, pediatric patients, and those with organ impairment—remain significant clinical challenges, with suboptimal dosing contributing to therapeutic failure or adverse drug events in an estimated 30–40% of treated patients worldwide.

This study presents a comprehensive data-driven framework for predictive analytics in precision drug delivery, integrating a hybrid physiologically-based pharmacokinetic-machine learning (PBPK-ML) model with multi-objective dose optimization. A real-world clinical dataset of 10,000 patient records encompassing six drug classes—antibiotics, anticoagulants, immunosuppressants, antiepileptics, oncology agents, and cardiovascular drugs—was compiled from electronic health records and therapeutic drug monitoring databases.

The proposed ensemble learning architecture, combining XGBoost, Random Forest, and a deep neural network through stacked generalization, achieved a mean absolute percentage error (MAPE) of 7.8% and R² of 0.956 for plasma drug concentration prediction across all drug classes. The PBPK-ML framework achieved clinically significant improvements in therapeutic target attainment (TTA) across seven patient subgroups, with the greatest gains observed in renally impaired patients (+35.5% TTA) and elderly patients (+30.6% TTA). Adverse drug event rates were reduced by a mean of 38.5% compared to standard weight-based dosing. SHAP-based interpretability analysis identified creatinine clearance, CYP450 genotype, and patient weight as the three most influential predictors of drug exposure variability.

These findings demonstrate the transformative potential of machine learning-enhanced precision dosing for improving clinical outcomes across diverse patient populations.

Keywords: Precision drug delivery, Predictive analytics, Machine learning, PBPK modeling, Therapeutic drug monitoring.

How to cite this article: Salunke G, Salunkhe SS, Mahajan K, Koban MS, Rana M, Mulani AO. Predictive analytics in precision drug delivery: a data-driven approach using machine learning. Int J Drug Deliv Technol. 2026;16(3s): 987-997; DOI: 10.25258/ijddt.16.3s.119

Source of support: None.

Conflict of interest: None